A data compliance sharing algorithm for intelligent connected vehicles empowered by federated learning

Zhou, Bei · 2025 · Crossref

DOI: 10.1504/ijict.2025.10073652

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the critical challenge of balancing data utility with privacy protection and regulatory compliance in Intelligent Connected Vehicle (ICV) data sharing. As ICVs generate massive amounts of sensitive data, traditional centralized sharing models pose significant privacy leakage risks and high transmission costs. The study proposes a federated learning (FL) framework that integrates differential privacy (DP) and Paillier homomorphic encryption (HE) to enable secure, compliant multi-party collaboration without exposing raw data. The proposed algorithm utilizes a four-layer architecture designed to ensure privacy, efficiency, and legal compliance. The data privacy protection layer combines DP, which adds noise to prevent individual identification, with HE, which allows computations on encrypted data. The FL layer employs federated averaging (FedAvg) for model aggregation, while the data transmission layer reduces bandwidth consumption through 8-bit gradient quantization, sparsification, and gradient compression protocols. The compliance validation layer implements role-based access control (RBAC) and real-time compliance auditing to ensure adherence to regulations like GDPR. Finally, a feedback layer uses momentum-accelerated averaging and adaptive communication frequency tuning to optimize model convergence and network efficiency. Experimental validation was conducted using two self-constructed datasets: the ICV-Behaviour-Env-Dataset (50,000 driving scenes) and the ICV-Sensor-Env-Dataset (20,000 environmental scenes). Results demonstrate that the integrated DP and HE approach achieves a model accuracy of 93.5% while reducing privacy leakage risk by 79% compared to the baseline (0.21 vs. 0.98). This combined method outperforms standalone DP (0.6 risk) and HE (0.5 risk) implementations, offering a superior privacy-utility balance. Additionally, the optimization techniques, including momentum acceleration and gradient quantization, reduced bandwidth consumption by 62%, addressing the high communication costs typical in ICV networks. The study concludes that integrating DP and HE within a federated learning framework provides a robust solution for ICV data sharing, effectively mitigating privacy risks while maintaining high model accuracy. The inclusion of compliance auditing and access control mechanisms ensures that data sharing meets legal standards, facilitating safer and more efficient intelligent transportation systems. This approach offers a scalable model for multi-party collaboration in autonomous driving, where data privacy and regulatory adherence are paramount.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• v2x connected vehicle